problem-solving

mway 的头像作者 mway

结构化问题解决方法。激活用于规划、设计、调试或复杂任务。用于系统分析和基于证据的推理。

📖 中文导读:结构化问题解决方法。激活用于规划、设计、调试或复杂任务。用于系统分析和基于证据的推理。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: problem-solving description: Structured problem-solving methodology. Activates for planning, design, debugging, or complex multi-step tasks. Use for systematic analysis and evidence-based reasoning. allowed-tools: ["shell", "read_file", "apply_patch", "update_plan"] metadata: short-description: Systematic problem-solving approach

Problem-Solving Methodology

Read these references:

  • references/methodology.md - 5-phase problem-solving framework (detailed)
  • references/behavior.md - Critical thinking and evidence-based reasoning
  • references/decomposition.md - Breaking problems into subproblems
  • references/parallelization.md - Identifying parallel execution opportunities

Instructions

Apply the 5-phase problem-solving framework from AGENT.md and references:

Phase 1: Understand (Decompose)

  • Gather all requirements and constraints
  • Identify unknowns and dependencies
  • Restate the problem to confirm understanding
  • Ask clarifying questions if needed
  • Read relevant code/documentation

Phase 2: Plan (Strategize)

  • Design approach from first principles
  • Make scope decisions (what's in/out)
  • Identify opportunities for parallelization
  • Break complex work into TODO items per task-management.md
  • Consider multiple solutions, pick best one

Phase 3: Execute (Implement)

  • Work through TODO list systematically
  • Update task status in real-time (exactly ONE in_progress)
  • Validate incrementally - don't save all verification for the end
  • Document decisions as you go
  • Apply critical thinking throughout

Phase 4: Verify (Validate)

  • Sanity check first: "Is this actually good/correct?"
  • Run mechanical verification (tests, lint, build)
  • Fix issues immediately, re-verify
  • Don't proceed until verification passes
  • Use evidence to confirm correctness

Phase 5: Reflect (Learn)

  • What worked well? What didn't?
  • Were estimates accurate?
  • Would a different approach have been better?
  • (This is internal reflection - share selectively)

All guidance from AGENT.md Task Tracking section also applies.